brucethemoose/CaPlatTessDolXaBoros-Yi-34B-200K-DARE-Ties

TEXT GENERATIONConcurrency Cost:2Model Size:34BQuant:FP8Ctx Length:32kPublished:Dec 9, 2023License:yi-licenseArchitecture:Transformer0.0K Cold

brucethemoose/CaPlatTessDolXaBoros-Yi-34B-200K-DARE-Ties is a 34 billion parameter language model based on the Yi-34B-200K architecture, created by brucethemoose. This model is a low-density DARE Ties merge, primarily intended for benchmarking on the open LLM leaderboard. It features a 32K context length and is noted by its creator as a lower-performing variant compared to a higher-density merge.

Loading preview...

Model Overview

This model, brucethemoose/CaPlatTessDolXaBoros-Yi-34B-200K-DARE-Ties, is a 34 billion parameter language model built upon the Yi-34B-200K architecture. It was created by brucethemoose using a low-density DARE Ties merging method, combining several base models including chargoddard_Yi-34B-200K-Llama, migtissera_Tess-34B-v1.4, bhenrym14_airoboros-3_1-yi-34b-200k, Nous-Capybara-34B, kyujinpy_PlatYi-34B-200K-Q, ehartford_dolphin-2.2-yi-34b-200k, and fblgit_una-xaberius-34b-v1beta.

Key Characteristics

  • Architecture: Based on the Yi-34B-200K model.
  • Parameter Count: 34 billion parameters.
  • Context Length: Supports a context window of 32,768 tokens.
  • Merge Method: Utilizes the DARE Ties merging technique with a low density configuration.

Intended Use

This specific model was primarily developed for benchmarking purposes on the open LLM leaderboard. The creator explicitly notes that this low-density merge is generally not recommended for general use cases, as a higher-density variant of the same merge is available and performs significantly better in both leaderboard scores and perplexity tests. Users seeking optimal performance are directed to the higher-density alternative.